[sergo] Sergo Report: DELTA-73to74-newlinter-audit(closeerrorunchecked)+reconcile-2-reverse-phantoms - 2026-09-30

[github-actions[bot]]

Executive Summary

Run 82 found the registry grew for the first time since R74 (73 -> 74 analyzers) with the addition of closeerrorunchecked, a linter that flags discarded Close() error return values. Its first-ever audit surfaced a fresh instance of the most recurring bug class in this repository, paren_unwrap_gap, on the very first day the linter existed. Reconciliation also caught two more reverse-phantom closures (issues auto-expired as not_planned while the underlying bug remained 100 percent present in code), continuing a long-running pattern where the issue tracker in this repository auto-expires stale-looking issues faster than they get fixed.

Strategy split this run: roughly 60% cached / 40% new-explore (leaning cached because a new linter appeared, which is itself the highest-value new-explore target by convention, so the new component was folded into the mandatory new-linter audit rather than a separate cold-start probe).

Tool Updates

• Serena: stable at 24 tools, activate_project succeeded for go/typescript/bash with no approval prompt.
• Registry: DELTA 73 -> 74. 74th analyzer = closeerrorunchecked (PR [linter-miner] Add close-error-unchecked linter for resource cleanup error detection #64310, merged 2026-09-29, one day before this run).
• closeerrorunchecked is not yet listed in .github/workflows/cgo.yml LINTER_FLAGS, so it is diagnostic-only, not CI-enforced, for now.

Strategy Split Details

Cached component (reconcile, ~60%): Re-verified all 15 issues open at the start of this run via gh api issues?labels=sergo&state=open. Two had auto-expired to closed/not_planned since the last run:

• bufioscannererunchecked: block-scope Err() check gap still unfixed (issue 60738 auto-expired, 7th reverse-phantom) #62541 (bufioscannererunchecked, block-scope Err() gap) - re-read analyzeBlockStatements/hasScannerErrCheck, confirmed the per-block-only forward search is still exactly as filed. This is the second time this specific bug lineage (60738 -> 62541 -> new) has been closed without a code change.
• uncheckedsliceindex: selector-based slice/string bases defeat all bounds-check and safe-loop recognition #62540 (uncheckedsliceindex, selector-base blind spot in sameExpr) - re-read lines 366-371, confirmed the Ident-only equality check is still unchanged, still leaves the two previously-identified production false positives in pkg/workflow/skills_ref_resolution.go and pkg/workflow/cache_memory.go unresolved.

Both were refiled with full evidence citations rather than treated as fixed.

New-explore component (~40%): With the registry growing for the first time in 8 runs, budget went to a full first-read audit of closeerrorunchecked rather than a cold probe of already-audited linters. The audit confirmed the linter is well-designed on type-identity (it correctly distinguishes the builtin error interface from a package-local type literally named error in its own test fixtures, avoiding the syntactic_stdlib_match class entirely) but reproduces the paren_unwrap_gap class that has now been found independently in roughly a dozen other linters across the history of this codebase.

Findings

1. closeerrorunchecked: parenthesized Close() discards escape detection (new, latent, not yet CI-enforced) - both patterns in analyzeAssignStmt do a bare rhs.(*ast.CallExpr) assertion with no astutil.UnwrapParenExpr first, so _ = (f.Close()) and value, _ := (closer.Close()) are invisible despite being semantically identical to the already-flagged bare forms.
2. bufioscannererunchecked: block-scope Err() gap (reverse-phantom refile, latent, not CI-enforced) - an Err() check placed in an enclosing block after the block containing the scan loop is invisible to the per-block-only forward search.
3. uncheckedsliceindex: selector-base blind spot in sameExpr (reverse-phantom refile, latent, not yet CI-enforced but affects 6 call sites and 2 confirmed live production sites) - struct-field-based slice/string indices are invisible to every bounds-check/safe-loop/terminating-guard recognizer in the linter because the shared equality helper only accepts bare identifiers.

Generated Tasks / Issues Created

Three issues filed this run (max allowed: 3):

• aw_sg82a1 - closeerrorunchecked paren_unwrap_gap (new finding, small effort)
• aw_sg82a2 - bufioscannererunchecked block-scope Err() gap (reverse-phantom refile, medium effort)
• aw_sg82a3 - uncheckedsliceindex sameExpr selector-base blind spot (reverse-phantom refile, medium effort, highest leverage: one shared-helper fix benefits 6 call sites)

Dedup was confirmed via gh api search/issues before filing all three (zero prior bug-report coverage beyond the closeerrorunchecked creation PR itself, and the two reconciled issues are direct refiles of their own prior filings).

Metrics

• Findings this run: 5 (3 filed as issues, plus the type-identity/customerror design confirmation and the not-yet-CI-enforced status check, both folded into the write-up rather than filed separately)
• Tasks generated: 3
• Issues created: 3 of 3 allowed
• Success score (self-assessed): 9/10 - a new linter appeared (highest-value signal), the audit found a real bug on day one, and both reconciliation refiles were fully code-verified with concrete evidence before refiling.

Historical Context

This is run 82 of an ongoing daily series (started ~R42 in the continuous history of this memory). Running totals after this run: 82 total runs, ~508 cumulative findings, ~146 cumulative tasks/issues, average success score ~8.75/10. The registry has grown from 42 analyzers (R58) to 74 today, with closeerrorunchecked as the 14th analyzer added since the registry migrated to pkg/linters/registry.go at R61. The paren_unwrap_gap class alone has now been found in at least 13 distinct locations (nilctxpassed, errstringmatch, sprintfbool, tolowerequalfold, timenowsub, httpstatuscode, seenmapbool, the shared astutil string-literal helpers, and now closeerrorunchecked), making it the single most recurring bug class this workflow has ever catalogued - strong evidence that a shared lint-authoring template or code-review checklist item (call astutil.UnwrapParenExpr before any bare AST type assertion on a call/literal/ident) would prevent an entire category of future first-audit findings.

Recommendations

1. Given how consistently paren_unwrap_gap recurs in brand-new linters (this is now confirmed in a linter that did not exist when the class was first catalogued), consider a repository-level lint-authoring guideline or shared helper wrapper that makes UnwrapParenExpr the default rather than an opt-in call developers must remember.
2. The reverse-phantom-reconcile pattern (auto-expire without fix) continues at a steady rate (this run: 2 of 15 open issues, roughly 13 percent) - if this is a GitHub auto-stale-close behavior rather than a deliberate not-planned decision by a maintainer, it may be worth exempting the sergo label from auto-expiry so genuine unfixed bugs stop needing manual re-filing every few runs.
3. The sameExpr fix in uncheckedsliceindex (issue aw_sg82a3) is the highest-leverage open item in the current backlog: a single shared-helper change would resolve findings across 6 call sites and 2 confirmed live production false positives at once.

Next-Run Focus (R83)

• Verify sg82a1/sg82a2/sg82a3 landed as filed.
• Re-tally open issues (expect 15: 12 unchanged plus the 3 filed this run).
• If the registry grows again (75th analyzer), audit it fresh as the primary new-explore target, as done this run.
• Check whether closeerrorunchecked gets added to cgo.yml LINTER_FLAGS (would make aw_sg82a1 CI-enforced rather than latent).
• Low-priority follow-up: the ident-only fallback path in closeerrorunchecked (hasCloseMethodReturningError via named.Methods()) does not include promoted/embedded methods - only relevant when pass.TypesInfo.Selections[sel] is nil, not yet confirmed with a live repro, worth a quick check if no new linter appears next run.

Generated by 🤖 Sergo - Serena Go Expert · claude · agent · 258.2 AIC · ⌖ 4.64 AIC · ⊞ 4.9K · ◷

• expires on Sep 30, 2026, 8:02 PM UTC-08:00

[github-actions[bot]]

This discussion has been marked as outdated by Sergo - Serena Go Expert.

A newer discussion is available at Discussion #64682.